Complete Theme Section in pdf format - Inter Research

250
Mar Ecol Prog Ser 459: 247–258, 2012
taining target levels of spawning stock biomass and
specifying corresponding fishing mortality rates
using spawning-biomass-per-recruit analyses. For
index-based assessments, we expanded survey
catch-per-tow estimates to total biomass using the
area-swept and survey catchability approach de -
scribed above. Once placed on an absolute scale,
these reference points could then be compared with
the production model results. Again, because the
basis for selecting and estimating reference levels
differed markedly from the production model ap -
proach, the GARM results are intended only to provide
a general point of comparison and to test for
concordance and directionality of overall results.
Environmental effects
We examined the potential effect of environmental
or climate-related factors on surplus production in this
system. The covariates used were the Atlantic Multidecadal
Oscillation (AMO) index, the winter (December
to February) North Atlantic Oscillation (NAO)
index, and the Extended Reconstructed Sea Surface
Temperature (ERSST) index for the Northeastern US.
Both the NAO and AMO are basin-wide indicators.
The AMO index is based on spatial patterns in SST
variability after removing the effects of anthropogenic
forcing on temperature (Enfield et al. 2001). The NAO
is the dominant mode of climate variability over the
North Atlantic Basin (Hurrell 1995) and is known to
exert important ecosystem effects (Stenseth et al.
2002). The broad-scale ERSST series is based on a
comprehensive analysis of long-term temperature
records obtained from ships-of-opportunity (Smith &
Reynolds 2003, 2004). In each case, we converted the
observations to standard normal deviates with 0 mean
and unit standard deviation.
To examine candidate lags to be included in the
analysis, we first examined the cross-correlation
structure (Box & Jenkins 1976) for ASP and each of
the environmental covariates. Following identification
of candidate environmental variables and lags,
we incorporated these metrics in an extended production
model:
∑ δ jX
jt , −τ
j=
1
ASP = ( α−βB
) B e + e
t t t
where δ j is a coefficient for the effect of the jth covariate,
X j,t-τ is the value of covariate j at lag τ , and all
other terms are defined as before. All candidate models
were compared using the Akaike Information Criterion
(AIC; Akaike 1992) to test whether the fit with
environmental covariates performed better than the
base case (Eq. 2). We used the corrected AIC c to
adjust for sample size effects:
2KK
( + 1)
AIC c =− 2log e ( L)
+
(7)
n−K
−1
where L is the likelihood estimate, K is the number of
parameters estimated, and n is the number of observations
(Burnham & Anderson 1998).
m
t
(6)
RESULTS
Fig. 2. Biomass (kt) of selected demersal fish species for the Gulf of Maine
during 1964 to 2007
Aggregate biomass of the 12-
species complex de clined through the
mid-1990s but has since in creased,
nearly recovering to the levels at the
start of the survey time series (Fig. 2).
Substantial changes in relative species
composition are again evident
throughout the series with important
changes in the abundance of key species
such as cod, haddock, and redfish.
The coefficient of variation of the
aggregate biomass series (30.5%)
was substantially lower than that of
the individual species, which ranged
from 50.0 to 87.6% (mean 64.8%).
The greater relative precision of the
combined series is a potential advantage
of dealing with aggregate data.